Optimal window size extraction for the Proper Orthogonal Decomposition technique applied to Wildfire Detection

A major environmental hazard is constituted by wildfires. In order to reliably detect them, remote sensing techniques are widely used to monitor eventual temperature anomalies. I apply the Proper Orthogonal Decomposition (POD) technique to thermal maps obtained from the Spinning Enhanced Visible and infrared Imager (SEVIRI) onboard the Meteosat Second Generation (MSG) satellite to detect thermal anomalies. I study a wildfire event from August 08th 2021 in Calabria, Southern Italy, for which both the coordinates and time of the fire were known. Middle Infrared (MIR, 3.9 μm) observations collected at 15-minute intervals over a whole day have been analyzed at different window sizes. This study is motivated by the fact that, being the POD a statistical technique, the size of the detection window can strongly influence the results. In fact, in the POD, the autocorrelation function of the data is used to capture the most energetic contributions to the signal. Because of this, a large window size of data can include too large environmental fluctuations that mask the thermal anomalies, while a too-narrow window implies a lower statistical significance of the dataset, decreasing the quality of the results.

POD has been employed here analyzing four different spatial extraction windows (3×3, 15×15, 30×30, and 45×45 pixels). The analysis focuses on higher-order POD modes, the 6th, 7th, and 8th modes, which show better detection capabilities of the wildfire activity. Results indicate that wildfire occurrence locations and times have been successfully detected by the POD and that the 15×15 pixel window represents the optimal balance among anomaly enhancement, spatial localization, and noise reduction. The POD capability for fire-detection is currently under investigation and will be subject of a forthcoming article for the journal Atmosphere. In this contribution, I investigated the window size that yields the best results in the fire identification.